Quantifying Weighted Morphological Content of Large-Scale Structures via Simulation-Based Inference

This study demonstrates that combining Minkowski Functionals with Conditional Moments of Derivatives (CMD) via simulation-based inference yields significantly tighter cosmological constraints on $\sigma_8$ and $\Omega_{\mathrm{m}}$ than either statistic alone, with the CMD-enhanced morphological approach outperforming the redshift-space halo power spectrum in mass-selected halo configurations by capturing complementary anisotropic and nonlinear features.

The Gist

Imagine you are trying to understand the shape of a giant, invisible cloud of fog that fills the universe. This "fog" is actually made of dark matter and galaxies, forming a cosmic web of clusters, filaments, and empty voids. Astronomers want to know exactly how this web is built because its shape tells us the secrets of the universe's ingredients: how much matter there is ($\Omega_m$) and how clumpy it is ($\sigma_8$).

For a long time, scientists tried to measure this cloud by looking at simple statistics, like counting how many pairs of galaxies are a certain distance apart. This is like trying to understand a complex sculpture by only measuring the distance between two specific points on it. It works okay, but you miss the big picture.

This paper introduces a new, smarter way to "see" the universe using Simulation-Based Inference (SBI). Here is the breakdown of their method and findings in simple terms:

1. The Problem: The "Likelihood" Trap

Traditionally, to learn from data, scientists need a perfect mathematical formula (a "likelihood") that predicts what the data should look like for any given universe. But the universe is messy and non-linear (like a tangled ball of yarn). Creating a perfect formula for this mess is nearly impossible. It's like trying to write a recipe for a soufflé that accounts for every single draft of air in the kitchen.

The Solution: Instead of writing a recipe, the authors built a super-smart AI (a neural network). They fed the AI thousands of simulated universes with known ingredients and asked it to learn the pattern. Once trained, the AI can look at real data and guess the ingredients without needing a perfect formula.

2. The Tools: Three Different "Eyes"

The team compared three different ways of looking at the cosmic web to see which one gives the best answer:

• The Power Spectrum (PS): This is the "old reliable" method. It's like looking at the fog through a standard camera lens. It measures how clumpy the fog is at different sizes. It's good, but it mostly sees the fog as a smooth, round blob. It misses the weird shapes.
• Minkowski Functionals (MFs): This is like looking at the fog through a 3D sculptor's eye. Instead of just measuring distances, it measures the shape of the fog: How much volume does it take up? How much surface area does it have? Is it curved like a hill or a valley? It captures the geometry and topology (the "connectedness") of the universe.
• Conditional Moments of Derivatives (CMD): This is the new, super-powered eye. Imagine the fog isn't just a shape, but a shape that is being stretched by invisible winds. In our universe, the motion of galaxies stretches the fog in specific directions (towards us or away from us). The "CMD" tool is special because it doesn't just look at the shape; it looks at the direction of the stretch and the intensity of the wind. It's like adding a compass and a speedometer to the sculptor's eye.

3. The Experiment: The "Big Sobol Sequence"

The authors used a massive library of 32,768 simulated universes (called the Big Sobol Sequence). They didn't just pick one "standard" universe; they simulated a huge variety of them with different amounts of matter and clumpiness. They trained their AI on 25,000 of these and tested it on the rest.

4. The Results: Who Won?

When they compared the tools, the results were surprising and exciting:

• Shape vs. Direction: The "Sculptor's Eye" (MFs) was better than the "Standard Camera" (Power Spectrum) at finding the universe's ingredients. But the "Super-Powered Eye" (CMD) was even better.
• The Power of Teamwork: When they combined the Sculptor's Eye (MFs) with the Super-Powered Eye (CMD), they got the best results of all. It's like having a sculptor who also knows how the wind blows. This combination improved the precision of their measurements by about 27% for the amount of matter and 26% for the clumpiness, compared to using just the shape tools alone.
• The "Massive Halos" Surprise: They tested what happens if they only look at the heaviest, most massive clumps of matter (ignoring the tiny ones).
  • The "Standard Camera" (Power Spectrum) got confused and its measurements got worse.
  • The "Super-Powered Eye" (CMD) actually got better (improving by 43% for matter density!).
  • Why? The heavy clumps are like giant, clear landmarks. The new method is so good at reading the direction and shape of these landmarks that it doesn't need the tiny, noisy details to get a perfect answer.

5. The Takeaway

This paper shows that to understand the universe, we need to stop just counting dots and start looking at shapes, directions, and how things are stretched.

By using AI to learn from simulations, the authors proved that a new type of measurement (CMD) combined with traditional shape analysis (MFs) can extract more information from the universe than the old methods. It's a bit like realizing that to understand a storm, you don't just need to measure the wind speed; you need to understand the shape of the clouds and the direction the wind is pushing them.

In short: They built a better "cosmic microscope" using AI, and it turns out that looking at the direction of the cosmic web gives us a much clearer picture of what the universe is made of.

Technical Summary

1. Problem Statement

Traditional cosmological inference relies on low-order summary statistics, primarily the power spectrum (PS) or two-point correlation functions. While computationally efficient in the linear regime, these methods discard significant information contained in the non-Gaussian and nonlinear evolution of the Large-Scale Structure (LSS). Furthermore, conventional methods require explicit analytical modeling of the likelihood function, which is often intractable for complex, higher-order observables.

With the advent of next-generation galaxy surveys (e.g., DESI, Euclid, Roman), there is a critical need for:

1. Advanced Statistics: Tools capable of capturing non-Gaussian signals and anisotropic features (specifically Redshift-Space Distortions, or RSDs) that scalar statistics miss.
2. Robust Inference Frameworks: Methods that can extract cosmological parameters without relying on explicit likelihood functions.

This work addresses the gap in quantifying the constraining power of weighted morphological descriptors compared to standard power spectrum multipoles, specifically focusing on their sensitivity to anisotropy and nonlinearity in redshift space.

2. Methodology

A. Data and Simulations

• Simulation Suite: The analysis utilizes the Big Sobol Sequence (BSQ) subset of the Quijote simulation suite.
• Parameters: 32,768 N-body simulations varying five cosmological parameters ($\Omega_m, \Omega_b, h, n_s, \sigma_8$) using a Sobol sequence for uniform coverage.
• Redshift: Analysis is performed at $z = 0.5$.
• Halo Catalogs: Generated using the Friends-of-Friends (FoF) algorithm. The study explores three halo selection scenarios:
  1. All halos (variable number density).
  2. Fixed number density ($\bar{n}_h = 10^{-4} h^3 \text{Mpc}^{-3}$).
  3. Mass-selected ($M > 3 \times 10^{13} h^{-1} M_\odot$).
• Redshift Space: Halo positions are converted to redshift space using peculiar velocities to incorporate RSDs (Kaiser effect and Fingers-of-God).

B. Summary Statistics

The study compares three classes of statistics:

1. Power Spectrum (PS): Redshift-space multipoles ($\ell=0, 2, 4$) up to $k_{\text{max}} \simeq 0.16 \, h \text{Mpc}^{-1}$.
2. Minkowski Functionals (MFs): Scalar morphological descriptors ($V_0, V_1, V_2, V_3$) quantifying volume, surface area, mean curvature, and Gaussian curvature of excursion sets.
3. Conditional Moments of Derivatives (CMD): A weighted morphological statistic introduced to capture directional/anisotropic information. CMD couples morphological measures with field derivatives (gradients), defined as:
   $$N^{(m)}_{\text{CMD}}(\vartheta, i) = \frac{1}{V} \int_V d^3x \, \delta_D(\delta_s(x) - \vartheta) \, |\nabla_i \delta_s(x)|^m$$
   This allows CMD to be sensitive to the orientation of anisotropies relative to the line of sight (LOS).

C. Inference Framework: Simulation-Based Inference (SBI)

• Approach: Likelihood-free inference using Neural Posterior Estimation (NPE).
• Model: Conditional Normalizing Flows (specifically Neural Spline Flows) trained to approximate the posterior distribution $p(\theta | D)$.
• Training: 25,000 simulations for training, 3,000 for validation, and 2,000 for testing.
• Calibration: Validated using simulation-based calibration (rank statistics) to ensure credible intervals are statistically accurate.

3. Key Contributions

1. Introduction of CMD in SBI: This is the first application of Conditional Moments of Derivatives (CMD) within a simulation-based inference pipeline for cosmological forecasting from halo fields in redshift space.
2. Quantification of Anisotropy Sensitivity: The study explicitly demonstrates that weighted morphological statistics (CMD) capture anisotropic information induced by RSDs that scalar MFs and standard PS multipoles miss or underutilize.
3. Comprehensive Benchmarking: A rigorous comparison of MFs, CMD, their combination, and the Power Spectrum across varying smoothing scales, fiducial cosmologies, and halo selection criteria.
4. Robustness Analysis: Investigation of how constraining power depends on halo abundance (number density) and mass resolution, revealing that weighted morphology is less dependent on tracer abundance fluctuations than the power spectrum.

4. Key Results

A. Performance of Weighted Morphology vs. Scalar Morphology

• CMD vs. MFs: At the fiducial cosmology ($\Omega_m = 0.3175, \sigma_8 = 0.834$) and smoothing scale $R=15 \, h^{-1}\text{Mpc}$, CMD provides systematically tighter constraints than MFs alone.
• Joint Estimator: Combining MFs and CMD improves precision by:
  • 27% for $\sigma_8$
  • 26% for $\Omega_m$
  • This highlights the complementary nature of scalar (geometric/topological) and directional (anisotropic) information.

B. Comparison with Power Spectrum

• Matched Scales: Comparing at $k_{\text{max}} \simeq 0.16 \, h \text{Mpc}^{-1}$ (equivalent to $R=15 \, h^{-1}\text{Mpc}$):
  • $\sigma_8$: The combined MFs+CMD estimator outperforms the full power spectrum multipoles ($\ell=0,2,4$) by ~47%.
  • $\Omega_m$: The power spectrum is slightly tighter, but the difference is not statistically significant given the uncertainties.
• Mass-Selected Halos: In the mass-selected configuration ($M > 3 \times 10^{13} h^{-1} M_\odot$), the weighted morphological estimator significantly outperforms the power spectrum for both parameters:
  • 45% improvement for $\sigma_8$.
  • 43% improvement for $\Omega_m$.
  • Interpretation: Mass selection suppresses shot noise and enhances the coherent bias of massive halos, amplifying the geometric/topological features captured by MFs+CMD.

C. Dependence on Smoothing Scale and Halo Selection

• Smoothing Scale: The constraining power of CMD is more sensitive to the smoothing scale $R$ than MFs. Smaller scales (higher $R$) yield tighter constraints for CMD because it captures non-linear information more effectively.
• Halo Abundance:
  • Fixing the number density degrades constraints for both PS and MFs+CMD, but the degradation is much larger for the Power Spectrum (54% loss for $\sigma_8$ vs. 30% for MFs+CMD).
  • This confirms that the power spectrum's constraining power on these scales relies heavily on abundance-driven information, whereas weighted morphology relies on the intrinsic geometry and anisotropy of the density field.

D. Stability Across Parameter Space

• While absolute uncertainties vary with the true cosmological parameters, the relative constraining power of the statistics (e.g., the ratio of uncertainties between MFs+CMD and PS) remains nearly constant across the explored parameter space.

5. Significance and Conclusion

This work establishes that weighted morphological statistics, particularly the combination of Minkowski Functionals and Conditional Moments of Derivatives, are superior to traditional power spectrum multipoles for extracting cosmological information from the nonlinear and anisotropic regimes of the LSS.

• Physical Insight: The results demonstrate that the "shape" and "direction" of cosmic structures (captured by CMD) contain distinct, high-value information about $\sigma_8$ and $\Omega_m$ that is not fully captured by clustering amplitudes alone.
• Methodological Advance: The study validates the use of SBI with complex, non-Gaussian descriptors, proving that likelihood-free inference can robustly handle these statistics without the need for analytical likelihood approximations.
• Future Outlook: The framework is robust against variations in tracer abundance and mass thresholds, making it highly promising for application to real-world galaxy surveys (like DESI and Euclid) where selection effects and systematics are significant. Future work will extend these comparisons to smaller scales ($k_{\text{max}} = 0.5 \, h \text{Mpc}^{-1}$) and incorporate observational systematics.